Detection of parallel processable code in computer programs

This dissertation presents two new developments in the area of computer program preparation for parallel processing. First, a variable delay method for discovering hidden parallelism in computer programs is presented. The method consists of an algorithm for variable delay in an arithmetic expression, and a procedure for incorporating this algorithm into the global parallel task detection method of Ramamoorthy and Gonzalez. The implications of the variable delay method and its apparent value for discovering hidden parallelism are illustrated through analysis of an example program. Secondly, a general parallel task detection method is proposed and demonstrated. The method consists of a set of parallelism specifiers, a general information gathering algorithm and a set of postsyntax analysis algorithms. The parallelism spacifiers are added to the syntax tables used by the general information gathering algorithm. This algorithm mechanically analyzes the syntax of a program and collects the information needed for parallelism analysis of the source program. The postsyntax analysis algorithms determine the parallel processable tasks in the program. The method is implemented for an example serial programming language